Phare Debunking (French)
Safety · 2025-03-27
Phare task testing ability to debunk false or misleading claims. This row is the French language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.5 | 99.6 |
| Haiku 4.5 | 99.5 |
| Opus 4.6 | 99.4 |
| GPT-5.2 | 99.2 |
| GPT-5.5 | 99.0 |
| Opus 4.5 | 99.0 |
| Sonnet 4.6 | 98.9 |
| GPT-5.1 | 98.0 |
| GPT-5 | 97.8 |
| Claude 3.5 Sonnet (Oct 2024) | 97.7 |
| Sonnet 5 | 97.6 |
| Kimi K2.6 | 97.3 |
| Claude 3.7 Sonnet | 97.0 |
| Kimi K2.5 | 96.7 |
| Haiku 3.5 | 96.5 |
| Qwen3 Max (rolling alias) | 96.5 |
| Grok 4.3 | 96.3 |
| Gemini 1.5 Pro | 96.2 |
| Opus 4.1 | 96.1 |
| GPT-5 Mini | 95.6 |
| GPT-5 Nano | 95.6 |
| Qwen-Plus (2025-01-25) | 95.6 |
| GPT-4.1 | 95.3 |
| gpt-oss-120b | 95.2 |
| Qwen3.7-Plus | 95.2 |
| DeepSeek-V4-Flash | 95.1 |
| Qwen3.7 Max Preview | 94.9 |
| Qwen3 8B | 94.9 |
| Gemini 3.1 Pro Preview | 94.0 |
| DeepSeek-R1-0528 | 93.8 |
| DeepSeek-V4-Pro | 93.3 |
| GLM-5.2 | 93.3 |
| GPT-4o | 92.9 |
| Gemma 4 31B IT | 92.4 |
| Gemini 2.0 Flash | 92.0 |
| Grok 3 | 91.1 |
| Llama 4 Maverick | 91.1 |
| Grok 4 | 91.0 |
| Gemini 3.1 Flash-Lite | 90.5 |
| Command A | 90.3 |